pstack: build-the-lever can be a skill your subagents follow (#91)

This commit is contained in:
lauren
2026-05-27 21:40:03 -07:00
committed by GitHub
parent 6f2e0c6318
commit 65f4dacee2
@@ -1,6 +1,6 @@
---
name: principle-build-the-lever
description: "Apply when work is repetitive or bulk: many similar edits, a check you'll rerun, a population to transform. Build the tool that amortizes it (codemod, script, generator) once you know the recipe, instead of grinding by hand."
description: "Apply when work is repetitive or bulk: many similar edits, a check you'll rerun, a population to transform, a fan-out to subagents. Build the tool that amortizes it (codemod, script, generator, or a skill your subagents follow) once you know the recipe, instead of grinding by hand."
disable-model-invocation: true
---
# Build the Lever
@@ -13,6 +13,7 @@ When the work repeats, build the tool that does it instead of grinding by hand.
- Do the first few by hand to learn the exact recipe, then build the tool. Don't build on a guess.
- Codemod or script for bulk edits, generator for repetitive files, a dump-to-sqlite query for repeated analysis, a rerunnable check for repeated verification.
- When you fan work out to subagents, write the lever as a skill they all read: the recipe, the verification contract, and the do-not-touch fences in one artifact, so every delegate inherits the same hardened version instead of re-explaining it per prompt and watching each one drift. Harden it the moment a delegate games or drops the contract, then re-dispatch. Keep it outside the delegates' write scope so they can't quietly edit the contract.
- Commit it when the work outlives the session, so the next run reruns it instead of redoing it.
**Balance:** Leverage, not gold-plating. The [Laziness Protocol](../principle-laziness-protocol/SKILL.md) still holds. Build the lever only when it pays for itself across the remaining work, never for a one-off.